Skip to content

fix: apply Claude thinking parameters from the evaluated LangChain config - #99

Merged
andrewklatzke merged 1 commit into
mainfrom
aklatzke/AIC-3400/fix-langchain-thinking-and-examples
Sep 17, 2026
Merged

fix: apply Claude thinking parameters from the evaluated LangChain config#99
andrewklatzke merged 1 commit into
mainfrom
aklatzke/AIC-3400/fix-langchain-thinking-and-examples

Conversation

@andrewklatzke

@andrewklatzke andrewklatzke commented Sep 17, 2026 •
edited by cursor Bot
Loading

Copy link
Copy Markdown
Contributor

Summary

  • Stop hardcoding Anthropic thinking budget and max tokens in the LangChain thinking example so flag model.parameters are applied after evaluation.
  • Align LangChain README factory examples so the model name wins over a colliding model key in the parameter bag.
  • Refresh uv.lock workspace versions to match the published package metadata.

Fixes AIC-3400.

Test plan

  • uv run python main.py langchain-thinking launch-darkly-documentation-summarizer-messages-claude "Reason it out yourself without any tools: what is 17 times 23?" exits 0 with a non-empty response and usage totals > 0
  • Same example with doesnt-exist exits 1 with a clean error and no new output/ JSON
  • uv run python main.py native-graph-langchain travel-agent-flow "Book me a flight to Paris" exits 0 with a non-empty response and usage totals > 0
  • Same example with travel-agent-flow-wrong-key exits 1 with a clean disabled-graph error and no new JSON

Note

Overview
Stops overriding Anthropic extended-thinking settings in the LangChain thinking example so ChatAnthropic is built from the evaluated flag’s model.parameters (plus the configured model name), instead of hardcoded thinking / max_tokens constants.

Updates LangChain agents and messages README factory snippets so the flag’s model name is applied after spreading model.parameters, ensuring it wins if parameters include a colliding model key.

Bumps workspace package versions in uv.lock (e.g. 0.2.1 → 0.2.2) to align with published metadata.

Reviewed by Cursor Bugbot for commit 408ad5b. Bugbot is set up for automated code reviews on this repo. Configure here.

@andrewklatzke @cursoragent
fix: apply Claude thinking parameters from the evaluated LangChain co…
408ad5b 
…nfig

Hardcoding thinking budget and max_tokens in the example collided with flag parameters and broke langchain-thinking integration runs.

Co-authored-by: Cursor <cursoragent@cursor.com>
@andrewklatzke
andrewklatzke merged commit 8be2318 into main Sep 17, 2026
8 checks passed
@andrewklatzke
andrewklatzke deleted the aklatzke/AIC-3400/fix-langchain-thinking-and-examples branch September 17, 2026 22:59
@github-actions github-actions Bot mentioned this pull request Sep 17, 2026
Merged
andrewklatzke pushed a commit that referenced this pull request Sep 22, 2026
@github-actions
chore: release main (#91)
520b7e3 
🤖 I have created a release *beep* *boop*
---


<details><summary>launchdarkly-ai-server: 0.2.3</summary>

##
[0.2.3](launchdarkly-ai-server-0.2.2...launchdarkly-ai-server-0.2.3)
(2026-09-21)


### Features

* **AIC-3106:** add multimodal history support to graph().invoke()
([#26](#26))
([c91132d](c91132d))
* **client:** stamp modelKey and modelVersion from _ldMeta on tracking
events
([ef103d3](ef103d3))
* **client:** stamp modelKey and modelVersion from _ldMeta on tracking
events ([#97](#97))
([4e6998a](4e6998a))
* **evaluations:** add LD judge event support
([#63](#63))
([ce61644](ce61644))


### Bug Fixes

* **client:** harden model stamps and keep judge results from inheriting
parent model identity
([5d1e678](5d1e678))
* **client:** only stamp a non-empty string modelKey
([8424452](8424452))
* **evaluations:** dedup criteria case-insensitively, matching the API
([4126d7b](4126d7b))
* **evaluations:** offline judge message_history must carry
FORMATTING_INSTRUCTIONS
([0b81f4e](0b81f4e))
* **evaluations:** prefer an exact-provider judge handler over a
wildcard
([c61b3a5](c61b3a5))
* **evaluations:** route judge configs to a compatible handler
([1a5e099](1a5e099))
* extract LangChain content-block text and apply model parameters after
eval ([#80](#80))
([a4e1aad](a4e1aad))
* **graph:** prefer node tools before synthetic handoff routing
([#90](#90))
([f830b2c](f830b2c))


### Documentation

* **evaluations:** document criteria, judge handlers, and scoring policy
([5f5a626](5f5a626))
* **evaluations:** use a customer-style judge key in examples
([ceca049](ceca049))
* remove the internal staging host from a public repo
([00b8dba](00b8dba))
* remove the internal staging host from a public repo
([#96](#96))
([1ad638d](1ad638d))
* replace stale launchdarkly-ai package name with launchdarkly-ai-python
([#40](#40))
([3256db7](3256db7))
</details>

<details><summary>launchdarkly-ai-python: 0.1.7</summary>

##
[0.1.7](launchdarkly-ai-python-0.1.6...launchdarkly-ai-python-0.1.7)
(2026-09-21)


### Features

* **evaluations:** add LD judge event support
([#63](#63))
([ce61644](ce61644))


### Documentation

* **evaluations:** document criteria, judge handlers, and scoring policy
([5f5a626](5f5a626))
* **evaluations:** use a customer-style judge key in examples
([ceca049](ceca049))
* remove the internal staging host from a public repo
([00b8dba](00b8dba))
* remove the internal staging host from a public repo
([#96](#96))
([1ad638d](1ad638d))
* replace stale launchdarkly-ai package name with launchdarkly-ai-python
([#40](#40))
([3256db7](3256db7))
</details>

<details><summary>launchdarkly-ai-claude-agents: 0.2.3</summary>

##
[0.2.3](launchdarkly-ai-claude-agents-0.2.2...launchdarkly-ai-claude-agents-0.2.3)
(2026-09-21)


### Features

* **AIC-3106:** add multimodal history support to graph().invoke()
([#26](#26))
([c91132d](c91132d))


### Documentation

* replace stale launchdarkly-ai package name with launchdarkly-ai-python
([#40](#40))
([3256db7](3256db7))
</details>

<details><summary>launchdarkly-ai-claude-messages: 0.2.3</summary>

##
[0.2.3](launchdarkly-ai-claude-messages-0.2.2...launchdarkly-ai-claude-messages-0.2.3)
(2026-09-21)


### Features

* **AIC-3106:** add multimodal history support to graph().invoke()
([#26](#26))
([c91132d](c91132d))


### Documentation

* replace stale launchdarkly-ai package name with launchdarkly-ai-python
([#40](#40))
([3256db7](3256db7))
</details>

<details><summary>launchdarkly-ai-openai-agents: 0.2.3</summary>

##
[0.2.3](launchdarkly-ai-openai-agents-0.2.2...launchdarkly-ai-openai-agents-0.2.3)
(2026-09-21)


### Features

* **AIC-3106:** add multimodal history support to graph().invoke()
([#26](#26))
([c91132d](c91132d))


### Documentation

* replace stale launchdarkly-ai package name with launchdarkly-ai-python
([#40](#40))
([3256db7](3256db7))
</details>

<details><summary>launchdarkly-ai-openai-messages: 0.2.3</summary>

##
[0.2.3](launchdarkly-ai-openai-messages-0.2.2...launchdarkly-ai-openai-messages-0.2.3)
(2026-09-21)


### Features

* **AIC-3106:** add multimodal history support to graph().invoke()
([#26](#26))
([c91132d](c91132d))


### Documentation

* replace stale launchdarkly-ai package name with launchdarkly-ai-python
([#40](#40))
([3256db7](3256db7))
</details>

<details><summary>launchdarkly-ai-langchain-agents: 0.2.3</summary>

##
[0.2.3](launchdarkly-ai-langchain-agents-0.2.2...launchdarkly-ai-langchain-agents-0.2.3)
(2026-09-21)


### Features

* **AIC-3106:** add multimodal history support to graph().invoke()
([#26](#26))
([c91132d](c91132d))


### Bug Fixes

* **AIC-3382:** support Bedrock configs in LangChain handlers
([#101](#101))
([04a67c7](04a67c7))
* apply Claude thinking parameters from the evaluated LangChain config
([#99](#99))
([8be2318](8be2318))
* extract LangChain content-block text and apply model parameters after
eval ([#80](#80))
([a4e1aad](a4e1aad))


### Documentation

* replace stale launchdarkly-ai package name with launchdarkly-ai-python
([#40](#40))
([3256db7](3256db7))
</details>

<details><summary>launchdarkly-ai-langchain-messages: 0.2.3</summary>

##
[0.2.3](launchdarkly-ai-langchain-messages-0.2.2...launchdarkly-ai-langchain-messages-0.2.3)
(2026-09-21)


### Features

* **AIC-3106:** add multimodal history support to graph().invoke()
([#26](#26))
([c91132d](c91132d))


### Bug Fixes

* **AIC-3382:** support Bedrock configs in LangChain handlers
([#101](#101))
([04a67c7](04a67c7))
* apply Claude thinking parameters from the evaluated LangChain config
([#99](#99))
([8be2318](8be2318))
* extract LangChain content-block text and apply model parameters after
eval ([#80](#80))
([a4e1aad](a4e1aad))


### Documentation

* replace stale launchdarkly-ai package name with launchdarkly-ai-python
([#40](#40))
([3256db7](3256db7))
</details>

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).

<!-- CURSOR_SUMMARY -->
---

> [!NOTE]
> **Overview**
> **Release Please** cut that bumps all Python AI SDK packages and
records their release notes—no runtime code changes in this diff.
> 
> **`launchdarkly-ai-server` 0.2.3** and **`launchdarkly-ai-python`
0.1.7** ship the accumulated work since the last release: **multimodal
conversation history** in `graph().invoke()`, **LaunchDarkly judge
evaluation events**, and **model identity** (`modelKey` / `modelVersion`
from `_ldMeta`) on tracking events, plus fixes for judge handler
routing, criteria dedup, offline judge formatting, graph tool vs handoff
ordering, and LangChain content-block / post-eval parameters.
> 
> Integration packages (**Claude/OpenAI/LangChain** agents and messages
at **0.2.3**) pick up the multimodal `graph().invoke()` behavior via the
core client; **LangChain** packages additionally note **Bedrock config**
support and **Claude thinking** parameters from evaluated configs.
Versions are updated in `.release-please-manifest.json`, each package’s
`pyproject.toml`, `__version__`, and `CHANGELOG.md`.
> 
> <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit
3a40793. Bugbot is set up for automated
code reviews on this repo. Configure
[here](https://www.cursor.com/dashboard/bugbot).</sup>
<!-- /CURSOR_SUMMARY -->

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@atornsii atornsii atornsii approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@andrewklatzke @atornsii